skills/levnikolaevich/claude-code-skills/ln-810-performance-optimization-coordinator

ln-810-performance-optimization-coordinator

SKILL.md

Paths: File paths (shared/, references/, ../ln-*) are relative to skills repo root. If not found at CWD, locate this SKILL.md directory and go up one level for repo root.

ln-810-performance-optimization-coordinator

Type: L2 Domain Coordinator Category: 8XX Optimization

Coordinates performance optimization by delegating to L3 workers: ln-811 (algorithm), ln-812 (query), ln-813 (runtime). Workers run in parallel when inputs are independent.


Overview

Aspect Details
Input Target file/module OR audit report (ln-650 output)
Output Optimized code with verification proof
Workers ln-811 (algorithm), ln-812 (query), ln-813 (runtime)

Workflow

Phases: Pre-flight → Analyze Input → Delegate → Collect → Verify → Report


Phase 0: Pre-flight Checks

Check Required Action if Missing
Target file OR audit report Yes Block optimization
Git clean state Yes Block (need clean baseline for revert)
Test infrastructure Yes Block (workers need tests for keep/discard)

MANDATORY READ: Load shared/references/ci_tool_detection.md for test/build detection.


Phase 1: Analyze Input

Input Sources

Source Detection Workers Activated
docs/project/persistence_audit.md ln-650 output exists ln-812 (query) + ln-813 (runtime)
Target file + function User-specified ln-811 (algorithm)
Full audit report All ln-65X sections present All three workers

Worker Selection

Condition ln-811 ln-812 ln-813
Target function specified Yes No No
ln-651 findings present No Yes No
ln-653 findings present No No Yes
All audit findings Conditional* Yes Yes

*ln-811 activated only if specific algorithmic inefficiency identified in audit.


Phase 2: Delegate to Workers

CRITICAL: All delegations use Agent tool with subagent_type: "general-purpose" and isolation: "worktree" — each worker creates its own branch per shared/references/git_worktree_fallback.md.

Delegation Protocol

FOR each selected worker:
  Agent(description: "Optimize via ln-81X",
       prompt: "Execute optimization worker.

Step 1: Invoke worker:
  Skill(skill: \"ln-81X-{worker}\")

CONTEXT:
{delegationContext}",
       subagent_type: "general-purpose",
       isolation: "worktree")

Delegation Context

Field Type Description
projectPath string Absolute path to project
auditReport string Path to persistence_audit.md (if applicable)
targetFile string Target file path (if applicable)
targetFunction string Target function name (if applicable)
options.runTests bool Run tests after optimization
options.runLint bool Run lint after optimization

Parallelism

Workers Can Parallel Reason
ln-812 + ln-813 Yes Different files, different fix types
ln-811 + ln-812 Depends Only if targeting different files
ln-811 + ln-813 Depends Only if targeting different files

Rules:

  • If workers target the SAME file, run sequentially (ln-811 first, then ln-812/813).
  • Dependent workers share branch: If worker B depends on worker A's output (e.g., ln-813 needs ln-812's query changes), launch worker B in worker A's branch — so B sees A's changes.

Phase 3: Collect Results

Each worker produces an isolated branch. Coordinator aggregates branch reports.

Worker Branches

Worker Branch Pattern Contents
ln-811 optimize/ln-811-{function}-{ts} Algorithm optimizations with benchmarks
ln-812 optimize/ln-812-{ts} Query optimizations
ln-813 optimize/ln-813-{ts} Runtime optimizations

Result Schema

Field Type Description
worker string ln-811, ln-812, or ln-813
status enum success, partial, failed
branch string Worker's result branch name
fixes_applied int Number of kept optimizations
fixes_discarded int Number of discarded attempts
details object Worker-specific report

Phase 4: Aggregate Reports

Each worker verified independently in its branch (tests, build, lint run by worker itself). Coordinator does NOT rerun verification or revert worker changes.

On Failure

  1. Branch with failing tests logged as "failed" in report
  2. User reviews failed branch independently

Phase 5: Report Summary

Report Schema

Field Description
input_source Audit report or target file
workers_activated Which workers ran
total_fixes_applied Sum across all workers
total_fixes_discarded Sum across all workers
build_verified PASSED or FAILED
per_worker[] Individual worker reports
algorithm_improvement Benchmark improvement % (ln-811 only)

Configuration

Options:
  # Input
  audit_report: "docs/project/persistence_audit.md"
  target_file: ""
  target_function: ""

  # Workers
  enable_algorithm: true
  enable_query: true
  enable_runtime: true

  # Verification
  run_tests: true
  run_build: true
  run_lint: true

  # Safety
  revert_on_build_failure: true

Error Handling

Recoverable Errors

Error Recovery
Worker timeout Log timeout, continue with other workers
Single worker failure Revert worker changes, report partial success
Build failure Revert last worker, re-verify

Fatal Errors

Error Action
No workers activated Report "no optimization targets found"
All workers failed Report failures, suggest manual review
Dirty git state Block with "commit or stash changes first"

References

  • ../ln-811-algorithm-optimizer/SKILL.md
  • ../ln-812-query-optimizer/SKILL.md
  • ../ln-813-runtime-optimizer/SKILL.md
  • shared/references/ci_tool_detection.md

Definition of Done

  • Input analyzed (audit report or target file/function)
  • Appropriate workers selected based on input type
  • Workers delegated with worktree isolation (isolation: "worktree")
  • Each worker produces isolated branch, pushed to remote
  • Coordinator report aggregates per-worker results (branch, fixes, status)

Phase 6: Meta-Analysis

MANDATORY READ: Load shared/references/meta_analysis_protocol.md

Skill type: optimization-coordinator. Run after all phases complete. Output to chat using the optimization-coordinator format.


Version: 1.0.0 Last Updated: 2026-03-08

Weekly Installs
30
GitHub Stars
202
First Seen
6 days ago
Installed on
gemini-cli30
github-copilot30
codex30
kimi-cli30
amp30
cline30